Weightwatcher

Latest version: v0.7.5.2

Safety actively analyzes 629639 Python packages for vulnerabilities to keep your Python projects secure.

Scan your dependencies

Page 7 of 7

0.6.2

Added
Q in details dataframe
raw_distances implemented as distances(..., method=RAW)
distances(): added method=EUCLIDEAN | CKA
unit test(s) for distances




[0.6.x (in progress on trunk)] - 2022-10-05
Added
examples directory with notebooks and README
start_ids=1 added to analyze(), describe(), SVDSmoothing(), SVDSharpness() for issue 65
tqdm to setup.py
matplotlib-inline to setup.py

Removed
watcher.analyze(..., normalize=...).
pyRMT


Changed
powerlaw.Fit() wrapped in method to supress stdout and stderr
analyze(vectors=True) is turned back on as default
setup.py to use metaclass , not import ww, to allow better handling of dependencies
details.append() changed to details=pd.contact([details, data]) 137
models.vgg11(pretrained=True) to models.vgg11(weights='VGG11_Weights.IMAGENET1K_V1')
same for resnet



Fixed
TPL and E_TPL fits use xmin_distribution='trunctated_power_law'
replaced np.linalg.eig and np.linalg.svd with scipy version because they are faster and more numerically stable
analyze(savefig=False) now works in all cases
Fixed minor issues: 65, 69, 72, 82
tests/test.py Test_TFBert.test_num_layers added (but 74 kept)

Page 7 of 7

© 2024 Safety CLI Cybersecurity Inc. All Rights Reserved.